RoboCup European Championship:: Report of the Amsterdam 2000 event

Robotics and Autonomous Systems 36 (2001) 59–66

RoboCup European Championship:Report of the Amsterdam 2000 event

Frans Groen a,∗, Wiebe van der Hoek b,c, Pieter Jonker d, Ben Kröse a,Hans Spoelder e, Stefano Stramigioli f

a Faculty of Science, University of Amsterdam, Kruislaan 403, 1098 SJ Amsterdam, The Netherlandsb Department of Computer Science, Utrecht University, 3508 TB Utrecht, The Netherlands

c Department of Computer Science, The University of Liverpool, Liverpool, UKd Department of Applied Physics, Delft University of Technology, Loventzweg 1, 2628 CJ Delft, The Netherlands

e Division of Physics and Astronomy, Vrije Universiteit, De Boelelaan 1081, 1081 HV Amsterdam, The Netherlandsf Department of Information Technology and Systems, Delft University of Technology, Mekelweg 4, 2628 CD Delft, The Netherlands

Abstract

The authors, as the local organizers of the RoboCup European Championship, present a brief, informal overview of thisevent, with the aim to archive the main results, but also as a report to support organizers of future RoboCup events in theorganization and coordination of such a rather complex tournament. © 2001 Published by Elsevier Science B.V.

Keywords: RoboCup; Organization of European Championship

1. Introduction

The Robot World Cup, RoboCup [1], is an inter-national initiative to foster AI and intelligent roboticsresearch by providing a standard problem, a soccergame, where a wide range of technologies can beintegrated and examined. The first Robot World Cup,RoboCup’97, was held in Nagoya, Japan, in August1997, and included the participation of more than40 teams. RoboCup’98 was held in Paris, in July1998, where more than 60 teams participated. In1999, the third Robot Soccer World Cup was held inStockholm, Sweden, and in August 2000 the fourth

∗ Corresponding author.E-mail addresses: [email protected] (F. Groen), [email protected](W. van der Hoek), [email protected] (P. Jonker),[email protected] (B. Kröse), [email protected] (H. Spoelder),[email protected] (S. Stramigioli).

edition of the games was held in Australia. Futurescheduled RoboCup events are: RoboCup 2001Seattle and RoboCup 2002 Japan. Sites for 2003 andafter are not decided yet.

From 28 May till 2 June 2000, the first Euro-pean RoboCup Championship was organized inAmsterdam. This event, initiated through a proposalat RoboCup 1999 in Stockholm by Frans Groen, wasa first attempt to organize the RoboCup competitionon a regional basis. As such, it served as a tool toprobe the feasibility of the organization of such anevent on a local scale.

From start, the objective was to organize a low-costscientific event accessible to all interested researchgroups in the region. The second objective was to at-tract the attention of students-to-be and interest themfor sciences. This objective has left its mark on theorganization as such: members from the academiahave been the driving force behind this event.

0921-8890/01/$ – see front matter © 2001 Published by Elsevier Science B.V.PII: S0 9 2 1 -8 8 90 (01 )00135 -X

60 F. Groen et al. / Robotics and Autonomous Systems 36 (2001) 59–66

Fig. 1. Organization.

Without the active support of their respective institu-tions, expressed in the efforts of the faculty-boards,conference organizers, members of the system labs,volunteering students and joined task forces comprisedof employees of the research groups involved, an eventlike this would be very hard, if not impossible, torealize.

A similar remark could be made about two spon-sors: SONY and NWO. At the right time, they pro-vided the organizing committee with the necessaryfinancial backup to take the ‘go ahead’ decision. This,and much more were enabling factors for the FirstEuropean RoboCup Championship. In this report, wesummarize our findings and evaluate what has goneright and wrong. We hope that this evaluation willprove useful to others and stimulate them to takeRoboCup in all its forms one step further. A detailedreport including all the match results is available athttp://www.robocup.nl.

2. Organization

Preparation for the event started almost a year in ad-vance. Since the organization can hardly be qualifiedas a standard one, the need for a manual which sum-marizes the experience of the past is large. Althoughthe manual of Sweden was of great help to us, it ar-rived somewhat late and therefore we had to reinventthe wheel again to some extent. We have translatedthe Sweden manual to our local situation (see [2]).

To optimize the efficiency of the organization, itwas split into two parts: (i) organizational and logisticaspects and (ii) RoboCup related aspects. For the workof category (i), a professional organization was used,the University of Amsterdam (UvA) Congress Bureau,work in the second field was done by the academia.We think that, in principle, this is a good separation of

activities and responsibilities though it is difficult todefine a clear boundary. The organizational structureis elucidated in Fig. 1.

Most critical for the organization was the fundraising and finding of a proper place for the event.The first aspect has been a constant worry. Regard-ing the venue, the University Sports Center of theUvA offered us access at a substantial discount. Thisplace was approved by Mr. Kitano during a visit.

The committee met on a bi-weekly base andchecked progress according to the timetable. It shouldbe stressed that the complexity of the setup and theoverwhelming number of things that needed to bearranged made it difficult to keep clear track of theprogress. The committee worked according to theprinciple that every person had integral responsiblefor some type of activity, e.g. the person responsiblefor the simulation had to take care of projectors, com-puter time for the simulation connectivity, etc. Thecongress bureau handled all logistic aspects includingnegotiating with firms, reservation of the hotels, han-dling of the registration and all money transactions.

Public relation work was done by a professionalfrom the Technical University of Delft. Proper han-dling of the press requires a kind of unique person whocan both answer in-depth technical question as wellas have ample media experience. To facilitate this, wehave set up a Question/Answer form [2]. Although wehave managed to cover all the media and get ample ofattention, we still feel that more could have been madeof it. An aspect which one should not underestimateis the importance of the web.

3. Event

The event covered a full week. A precise planning isgiven in the timetable at [2]. As an evaluating remark,

F. Groen et al. / Robotics and Autonomous Systems 36 (2001) 59–66 61

Table 1Small-sized participants

Team Country Web

5DPO Portugal http://www.fe.up.pt/∼robosoc/CFA UPMC Paris France http://www.robo.jussieu.fr/rbo/Robocup/UPMC-CFA robocup team.htmFU Fighter Germany http://www.inf.fu-berlin.de/∼robocupGenius TaiwanRogi Spain http://rogiteam.udg.es/robots/rogi2team.htmlVUB AI-lab Belgium http://www.ida.liu.se/ext/epa/cis/1999/006/14/

we mention that the testing of the setup (transitionbetween Sunday and Monday) was maybe planned alittle tight.

3.1. Small sized

Six teams entered for the competition in thesmall-sized league. They were all placed within onepool. The teams are given in Table 1. This competi-tion was won by the FU Fighters, who did beat theRogi team with 6–0 in the finals. Third was the VUBAI-lab, fourth 5DPO.

3.2. Middle sized

In the middle-sized league, 10 teams competed. Theworld champion of 1999, the team Sharif, was alsoinvited to this European tournament. The arrival ofthe Sharif team was delayed, due to late subscriptionand (hence) visa problems. Consequently, up until alate state it was not certain whether there were 9 or10 teams for the competition. Most teams brought amodest number of team members (5–10) and robots

Table 2Middle-sized participants

Name Country Web

5DPO-2000 Portugal http://www.fe.up.pt/∼robosoc/Art Italy http://www.ai.sri.com/∼iocchi/robocup99/art.htmlAttempto Germany http://www-ra.informatik.uni-tuebingen.de/forschung/robocup/welcome.htmlDutch team The Netherlands http://www.robocup.nl/clockwork/index.htmlGMD Germany http://ais.gmd.de/BE/robocup/IsocRob Portugal http://socrob.isr.ist.utl.ptSharif Iran http://linux.ce.sharif.ac.ir/robocup/U. Minho Portugal http://www.robotica.dei.uminho.pt/robocup/Ulm Sparrows Germany http://smart.informatik.uni-ulm.de/SPARROWS/Uppsula Sweden http://www.docs.uu.se/robocup/2001/

along (3–5), however both the Swedish and Italianteams brought a considerable greater amount to thetournament, which gave a higher pressure on the avail-able space and infrastructure than was originally ex-pected. The teams are presented in Table 2.

The competition was set up as follows: 10 teamsplayed a half competition in two pools. The first eightentered a tournament — the quarter finals. The fourwinners of the quarter finals went into the semifinals.The two losers of the semifinals went for places 3 and4, the small finals. The two winners of the semifinalswent into the finals. We have tried to make the twoinitial pools of equal strength. We used the Stockholmrating for this.

Concerning technical issues, a miss was that someof the chairs were bright red, being cumbersome forsome robots that could look over the field boardings.The illumination was 1000 Lux uniformly distributed.This was sufficient. However, in the blue goals forthe Iranian team, more light had to be installed. Thepower for the paddock area was planned to be suffi-cient, however, the lighting company initially did notbring sufficient power distribution cables along, and


tapped in the team setup weekend all current fromthe (220 V) illumination net instead from the (380 V)powernet. This was solved on Monday when a (380to 280 V) converter was brought along (as was in theplanning) and more cables. All games have been tapedby Einstein TV using top cameras and some goalcameras, albeit that the quality is modest. Since theend of October the tapes are in Delft, where VHScopies can be ordered ([email protected]).

The colors that were used in Amsterdam were infact the colors used in Stockholm: Pantone ProcessBlue for the Blue Goal, Pantone Process Yellow forthe Yellow Goal, Pantone Process Cyan and PantoneProcess Magenta for the team shirts. The ball is (prob-ably) Pantone Orange 021. Pantone Process colors arethe basic colors used by printers to print color prints.The green of the corners is probably Pantone 340. Afew meters of fabric in magenta and cyan as shirtsor skirts for all the teams were available, however,hardly used, due to a new rule that team colors werenot mandatory.

On the Sunday before the games, a meeting aboutthe rules was organized. It became clear that severalteam leaders were not aware of what the rules actu-ally involved, or what the status was of the changesproposed on the web, i.e., what was ratified and whatwas only proposed. Moreover, it appeared that the def-inition of charging was unclear. The rules that we fi-nally agreed on had to be propagated to the teams bythe team leaders. Something that had not always been

Table 3Simulation league participants

Name Country Web: http://

Airg Sibiu Ulbs Romania airg.verena.ro/new/projects/AT Humboldt 2000 Germany www.ki.informatik.hu-berlin.de/Cyberoos 2000 Australia www.cmis.csiro.au/Essex Wizards Great Britain privatewww.essex.ac.uk/∼kkosti/FC Portugal Portugal www.ieeta.pt/robocup/Karlsruhe Brainstormers Germany www.karlsruhe-brainstormers.de/Mainz Rolling Brains Germany www.informatik.uni-mainz.de/ANGEW/Lucky Lubeck GermanyNo AI SwedenPizza Tower Italy medialab.di.unipi.it/Polytech 100 RussiaOULO2000 FinlandRoboLog Germany www.uni-koblenz.de/∼fruit/Wroclaw Poland www.ict.pwr.wroc.pl/robocup/

done or was not always digested by the team mem-bers, as we noticed later on during the games. More-over, not all dealmakers had already arrived in thatstage. Reason that the main rules were projected onthe wall during the games (also as a service to theaudience).

In general, the atmosphere in Amsterdam was goodand the competition was fair. Three new teams, fromHolland, Sweden and Portugal entered the league. Iranwon the games in a severe struggle with Italy. We thinkthat the final ranking reflected the strengths of theteams. Some progress with respect to Stockholm wasmade, in general better collision avoidance was seen,as well as better vision systems. Encouraging teambehaviors have been spotted. We propose not to leavethe walls out in future games, as the vision systemswill probably not be able to handle this. We would liketo enforce better color recognition and team behavior,as well as colored advertisements on the walls to startwith, combined with a larger field.

3.3. Simulation

All in all 14 teams (see Table 3) entered the com-petition for the simulation league. The teams weregrouped according to strength into two groups.

Airg Sibiu Ulbs did not show up, and PSI had trou-bles getting into the Netherlands and arrived too lateto effectively take part in the competition. The com-petition scheme was based on first a pool of which the


top four qualified and then a knock out scheme. Thecompetitors insisted on getting a full ranking of all theteams. FC Portugal was the winner in this league, Karl-sruhe Brainstormers was second, and the Essex Wiz-ards did beat Cyberoos 2000 for the third place: 2–1.

Concerning technical issues, contrary to other yearscompetitors were given one month in advance accessto the target system to compile and test their code. Thesimulation as such was run on a parallel system, theASCI DAS system, a distributed supercomputer. TheVrije Universiteit part of this machine contains 128nodes. For the matches, each player was mapped onto aprocessor. An extra processor was reserved for the soc-cerserver. To minimize the cross talk of network, trafficplayers of the two teams were separated by a switch.The system on which the simulation was run was con-nected to the standard visualization via an ISDN line.The bandwidth of this line was more than sufficient togive the updates for the visualization. Besides that, a3D visualization was run on an ImmersaDesk that wasplaced on the campus. Logfiles can still be retrievedfrom http://www.cs.vu.nl/das/robocup.logs/.

The parallel system proved to be very stable. Onlyin two cases was a restart of the match necessary. Thefact that the simulation is run on a remote system withproven stability adds much to decrease the runtimeproblems. During all the matches, traces of the net-work traffic induced have been taken. These will beanalyzed soon. Based on this analysis, we will try tosee if suggestion for an optimal configuration can bemade. It is clear already that the simulation is domi-nated by the network. Almost all competitors appreci-ated the possibility to test their code in advance verymuch. During the matches, almost no software prob-lems occurred.

3.4. Workshop

The workshop was scheduled in one full-day pro-gram on Wednesday 31 May, located at the Free Uni-versity. During this day, 14 papers were presented, giv-ing rise to lively discussions. The titles of the papers,together with the authors, are given below. Ranked tothe first author, they were from the following coun-tries: Germany (4), Italy (4), Portugal (2), Romania(1), Spain (1), United Kingdom (1) and Yugoslavia(1). A CD-ROM containing the papers was distributedat the workshop. Apart from the accepted papers, this

CD also contained 16 team descriptions: one for thesmall-sized league, 10 for the middle-sized, and fivefor the simulation league. Copies of this CD are stillavailable via [email protected].

Here is the list of accepted papers: 1

1. Simulation league1.1. Mihal Badjonski, Kay Schröter, Jan

Wendler and Hans-Dieter Burkhard, Learn-ing of Kick in Artificial Soccer.

1.2. Ciprian Candea, Marius Staicu and BoldurBarbat, Holon-like Approach for RoboticSoccer.

1.3. M. Riedmiller, A. Merke, D. Meier, A. Hof-mann, A. Sinner, O. Thate and Ch. Kill,Karlsruhe Brainstormers 2000 — DesignPrinciples.

2. Vision-based methods2.1. Stefan Enderle, Marcus Ritter, Dieter Fox,

Stefan Sablatnög, Gerhard Kraetzschmarand Günther Palm, Vision-Based Localiza-tion in RoboCup Environments.

2.2. Carlos Machado, Sergio Sampaio and Fer-nando Ribeiro, Image Processing Applied toa Robotic Football Team.

2.3. Fabio Marchese and Domenico Sorrenti,Omni-Directional Vision with a Multi-PartMirror.

2.4. Carlos Marques and Pedro Lima, A Local-ization Method for a Soccer Robot Using aVision-Based Omni-Directional Sensor.

2.5. Michael Plagge and Andreas Zell, Vision-Based Goal Keeper Localization.

3. (Team) behavior3.1. Claudio Castelpietra, Luca Iocchi, Daniele

Nardi and Maurizio Piaggio, Coordina-tion among Heterogenous Robotic FootballPlayers: ART in the F-2000 League.

3.2. Matthew Hunter, Kostas Kostiadis and Hu-osheng Hu, A Behavior-Based Approachto Position Selection for Simulated SoccerAgents.

3.3. Hans-Ulrich Kobialka, Peter Schöll andAnsgar Bredenfeld, Tools for AssessingRoboCup Behavior.

1 The final program of the workshop as well as a copy of thepapers can be found at [4].


3.4. Maurizion Piaggio and Antonio Sgorbissa,Exploiting ETHNOS for Communicationand Coordination of Heterogenous SoccerRobots in the ART Team.

4. Miscellaneous4.1. G. Adorni, S. Cagnoni and M. Mordonini,

Design Issues for the ART F-2000 RoboCupGoal-Keeper.

4.2. Josep de la Rosa, Bianca Innocenti, AlbertOller and Albert Figuerras, An Example ofDynamical Physical Agents.

The workshop was attended by almost all the partic-ipants discussing aspects of their tactics. In the currentedition, it is almost limited to the mid-sized league.

3.5. RoboCup Junior

During RoboCup’99 in Stockholm the firstRoboCup Jr. was arranged by Henrik Lund and LuigiPagliarini. The aim was to allow children get experi-ence with the programming of robots. A setup usingLEGO robots and a simple and clear programmingenvironment was made. During the preparation ofthe championship in Amsterdam, we noticed thatthere was a large interest of children for robot soccer.Therefore, we decided to organize RoboCup Jr. inAmsterdam, using a setup similar to the RoboCup Jr.1999 using LEGO robots and the software of Lund’sgroup. We had 12 teams (from the Netherlands andtwo from Germany) playing a real tournament.

The robot is based on a design by Lund of Lego-Lab Denmark. The basis is a standard LEGO Mind-storms set. The soccer robot is programmed usingthe ‘ILF’-software developed by Lund and Pagliariniwhich provides children an easy to use programmingenvironment. For a more detailed description of thehardware and software, we refer to [3]. The robotsplay in a one against one fashion. The field for the soc-cer game is a gray scale field printed on an oversizeA0 (119×87 cm2), surrounded by a wooden frame ofabout 13 cm high. The intensity level of the field formsa gradient from white to black, and using a light sen-sor underneath the robot, one can navigate around thefield. Before starting, the system has to be calibrated.

In contrast with the middle-sized league inRoboCup, the LEGO robots do not have a cameraon-board. This would require too much processing

Fig. 2. Robot with ball.

time and would result in a too expensive system. In-stead, the system uses LEGO light sensors, implyingthat an ‘active’ ball is needed. In RoboCup Jr. 2000,we used the balls which were designed in the LEGO-lab of the University of Aarhus (see Fig. 2). Theseballs contain LEDs which emit infrared light of awavelength which is optimal for the LEGO sensors.The ball was made of transparent plastic in which20 infrared emitters were positioned in a hexago-nal grid, so that coverage of all angles is assured.These balls were hand made, but a commercial ver-sion will be available soon from a Japanese companyElekit.

There were 12 teams from eight schools partici-pating in this tournament, in total about 50 childrenvarying from 13 to 16 years old. In the morning, thewhole group of participants was given a short generalintroduction after which every team was assigned to acomputer and a robot. We had one supervisor per threegroups, which is really needed. The younger childrenused the ‘beginners’ mode of the ILF program, whilethe older ones were encouraged to try the intermedi-ate level, in which they could also use conditionalsin their program. Despite the initial fear of program-ming, after 15 minutes everybody was very enthusias-tic. The first tests were carried out to investigate theeffect of the parameters of the program on the behav-ior of the robot. After one hour, we started playing thequalification rounds: from the 12 teams only 8 couldbe admitted in the quarter finals. After these qualifi-cation games, we still had teams of young children aswell as older ones in the tournament.


Table 4Winning team’s code

Der Manschaft31 mei 2000 14:58ua.ilf

SCORE IN WHITERobot reacts to Bumpers ActivityRobot Eyes Look at the Ball–

Beh:1 Find the ballBeh:2 Make a sound: BEEP BEEPBeh:3 Search and go to the ball: 5 sec.Beh:4 If robot sees the Ball CloseBeh:5 Go lighter: 6 sec.Beh:6 END If robot sees the Ball CloseBeh:7 If Robot is on White SideBeh:8 Go darker: 5 sec.Beh:9 END If White Side

After the qualification games, there was a lunchbreak and a demonstration of the ‘real’ RoboCupgames. We noted that some children wanted to con-tinue programming to improve their chances on win-ning. However, to have a fair tournament (and to havetime to recharge the balls), we imposed a strict onehour break. After lunch, we played the quarter finals,semifinals and finals on a single field in front of thesupporters. A LEGO Mindstorms set was given as afirst prize to the winners (Montessori Lyceum Ams-terdam). To our surprise, the second prize winner wasa group with a very simple program. The strength ofthis program is the speed it loops such that it is al-ways faster near the ball than complex programs. Thewinning team had a program which was developed inthe intermediate level mode, in which it was possibleto use IF-statements (in Table 4, Dutch commandsare translated into English).

The RoboCup Jr. games with the LEGO robots pro-vide a very good platform for the AI education ofchildren in the age between 8 and 15. The fact that aprogram has to be designed which controls a real sys-tem, and that this system has to compete with othersystems (of which at the moment the physical capabil-ities are identical) motivates the students enormously.At the moment, we are making an inventory of othersystems which have the same objectives. There is anincreasing interest in this field of edutainment, bothfrom industry as well as from academia. We believe

that the AI community will benefit from a good ex-planation of the problems and education in secondaryschools.

4. Conclusion

Organizing RoboCup, even on the scale at whichwe did it, and given the limited objective that we had,is a difficult task. It would have been impossible with-out the help of the local and international sponsors.That casts a shadow of a doubt on the feasibility ofRoboCup as such since also the participants have toinvest substantial amounts of money.

The simulation league is easy to organize and caneven be done in a virtual way: competitors submit theircode and watch remotely what is happening. In ouropinion, this possibility should be explored more ex-plicitly. The mid-sized league is very capital intensiveboth for the organizers as well as for the participants.We think that in the long run this cost effectivenesswill be a critical success factor for the competition.Similar arguments apply, albeit somewhat less, to thesmall-sized league.

We have tried to organize the first RoboCup Euro-pean Championships on a low budget basis. Realiza-tion of this is critical in our opinion for the survivalof RoboCup as an initiative. In retrospect, one can saythat we have succeeded in doing so, however, that thenet result is still mixed. It is doubtful whether organiz-ing RoboCup separate events — and attending themas participant — is a sustainable activity. One shouldstress that these conclusions do not hold for the sim-ulation: they are probably the most feasible activity.

Public attention focuses on the mid-sized league.To some extent, this is understandable because intu-itively these robots are the most appealing. Having saidthat, one should also point out that a number of the‘non-scientific’ visitors were somewhat disappointedin the state of the art of robotics. The simulation leaguegets little attention. This is surprising since the play isof exceptional high level in this league. Limiting factorhere is probably the restricted nature of the interface:people used to play with Fifa 99 are less impressed bya two-dimensional UI.

All in all, RoboCup is an investment in the futurebut a risky one: the return on investment in the shortterm is limited.


References

[1] http://www.robocup.org.[2] http://www.robocup.nl/robocup2000/.[3] carol.wins.uva.nl/∼krose/robocup jr/.[4] http://www.cs.uu.nl/∼wiebe/Robocup/.

Frans Groen is a Professor at the Uni-versity of Amsterdam and Head of the In-telligent Autonomous Systems Group. Heis a trustee of the International RoboCupFederation. His current research interestsare sensor data processing for autonomousmobile systems and shared dynamic worldmodeling.

Wiebe van der Hoek is a senior researcherat the Computer Science Department ofthe University of Utrecht and Professor atthe University of Liverpool. His currentinterests are logics for AI, belief revisionand agent programming languages.

Pieter Jonker is an Associate Professor inthe Pattern Recognition Group of the De-partment of Applied Physics, Delft Uni-versity. His current interest is in a projecton ubiquitous communications and nanocomputing.

Ben Kröse is an Associate Professor atthe Computer Science Institute of the Uni-versity of Amsterdam. His research in-terests are learning systems which per-ceive and act and computational intelli-gence methodologies for these systems. Heis member of the IEEE.

Hans Spoelder is a Faculty member of thePhysics Applied Computer Science Group,Division of Physics and Astronomy, VrijeUniversiteit, Amsterdam. His research in-terests include virtual realities, virtual en-vironments, virtual and distributed instru-mentation and man–machine interaction.

Stefano Stramigioli is an Associate Pro-fessor at the Control Laboratory of theFaculty of Information Technology andSystems, Delft University. He is currentlyinterested in modeling rigid bodies mech-anisms, interactive physical control andgrasping strategies for multilimbed sys-tems.

Documents

RoboCup European Championship:: Report of the Amsterdam 2000 event